The talent retention playbook that worked in 2015 just got stress-tested at AI escape velocity — and failed.
The Summary
- Thinking Machines Lab lost 13 of its 42 founding team members — nearly a third — in its first year, including three of six co-founders
- The exodus coincided with the one-year equity cliff, when early employees could cash out their first stock options
- Meta, OpenAI, and xAI are outbidding even well-funded AI startups for scarce technical talent
- The company scaled from 42 to 150+ people in a year, but growth doesn't mean retention when rivals throw money
The Signal
Mira Murati left OpenAI to build Thinking Machines Lab with $2 billion in backing and the kind of founding team that makes investors salivate. Deep ChatGPT training experience. OpenAI alumni. Technical chops that warranted a valuation before shipping a single product. One year later, a third of that founding crew is gone.
The timing tells the story. Most startup equity packages include a one-year cliff — you get nothing if you leave before twelve months, then a chunk vests all at once. That cliff just hit. Suddenly, people who joined for the mission and the upside had liquid options in hand and Meta knocking with offers rumored to be in the high seven figures for senior ML researchers.
"The one-year cliff is when startup compensation turns from promise into poker hand."
This isn't about Murati failing to inspire or Thinking Machines lacking ambition. It's about the math of the agent economy eating its own. When Meta needs 500 researchers to keep pace with OpenAI's roadmap, and OpenAI needs 300 to stay ahead of Anthropic, and xAI needs 200 because Elon won't accept second place — where do they get them?
They poach. From each other. From the newest labs. From anywhere signal-to-noise ratios are high.
The irony: Thinking Machines was supposed to be the destination, not the stepping stone. Murati's OpenAI credibility was the magnet. Early employees took below-market cash salaries betting the equity would print. Then the market for their skills moved faster than the equity could mature. A staff ML engineer at Meta now pulls $800K total comp. A founding engineer at a one-year-old startup with illiquid shares? Maybe $250K cash and paper worth... who knows.
Key dynamics at play:
- AI talent supply hasn't scaled with demand — still ~10,000 people globally who can actually build frontier models
- Big Tech has infinite capital and shipping products; startups have vision and equity that might be worth something in three years
- The vesting cliff creates a synchronized exit window that competitors exploit ruthlessly
The 150-person headcount growth shows Thinking Machines is hiring aggressively to backfill and scale. But there's a difference between filling seats and retaining the people who understood the original technical vision. Founders leave for different reasons than employees. When three of six co-founders bounce, that's not just about money — that's about direction, culture, or belief in the outcome.
The Implication
If you're building in AI right now, your retention strategy can't just be "we're going to be huge someday." The one-year cliff is a pressure test. Competitors will time their poaching to it. Your best people will have term sheets in hand the week they vest.
The answer isn't higher cash comp — you'll lose that arms race to Meta. It's faster value creation. Can you ship something in year one that makes the equity feel real? Can you hit milestones that justify a markup round before the cliff hits? Thinking Machines raised billions but hasn't publicly shipped. That's a long time to ask people to believe.
For talent: the best time to join an AI lab might be right after the one-year exodus. Team's been stress-tested, dead weight cleared out, and whoever's left actually wants to be there. Just make sure you're not walking into round two of the same problem.